David C. King

Harvard University

Richard E. Matland

The University of Houston

October 1, 1999

We thank Barbara Burrell, Rich Cooper, Debra Dodson, Anna Greenberg, Paul Gronke, Christine Matthews, Pippa Norris, Amber Ramage, Adrian Shepherd, Robin Worth, and Karen Jones Roberts for contributing to our thinking about this paper. Data were made available by RENEW, the Republican Network to Elect Women, and by the Center for the American Woman and Politics. Questions regarding the data should be directed to David_King@harvard.edu.

We report the results of an experiment involving 820 randomly sampled adults. Half heard about a female Republican candidate for Congress. The other half learned of an otherwise identical male candidate. A candidate’s gender clearly affected evaluations, but the effects of gender varied with the respondents’ party identification. Democratic and Independent voters were more likely to trust, think qualified, view as a leader, and vote for the female Republican (contrasted with the male Republican). We show that being female helps ease the predispositions Democrats and Independents have against Republican candidates. On the other hand, being female leads to associations that may hurt Republican women within their own party. Our analysis helps explain why female Republican candidates in the 1990s have been less likely to win their own party primaries, despite the expectation that they will do better than male Republicans in general election contests.

Does gender affect the way citizens evaluate candidates running for the United States Congress? Are there differential effects across the major parties in how voters assess female candidates? On the first question, recent scholarship highlights conditions under which females have certain advantages (Kahn 1996, Kahn & Gordon 1996, Iyengar et al. 1996, Burrell 1994). When campaigns evoke "female themes," such as education and social policy, voters apparently view women more favorably, and the news media tend to oblige by reinforcing sex-role stereotypes (Kahn 1992). Likewise, when campaign themes evoke issues that are stereotypically associated with males -- such as leadership, defense and economics -- female candidates are presumably disadvantaged (Fox 1997, Huddy 1994, Huddy & Terkildsen 1993a, Leeper 1991, Rosenwasser & Seale 1988, Sapiro 1981-82). Furthermore, some scholars argue that women are advantaged by an "affinity effect," in that females are more likely to vote for female candidates, and a slight majority of voters are female (Bernstein 1997; Garrett & Brooks 1987; Paolino 1995; Seltzer et al, 1997; Sigelman & Sigelman 1982; Spohn & Gillespie 1987; Thompson & Steckenrider 1997). In this paper, we investigate how electoral context – primarily partisanship –affects how a candidate’s gender influences voter opinions.

In some respects, the electoral context for female candidates as we enter the 21^st century seems especially promising. There are a record number of women in Congress (65 in 1999 compared to 25 in 1986) and a record number of women (89 in 1999) hold statewide executive posts. Over 40 percent of the Washington state legislature is female. In Arizona, women hold all five statewide elective positions. Nonetheless, there are signs that progress is coming very slowly. The percent of state legislators that are female stood at 20.5% in 1993 and has only risen to 22.3% by 1999. There are only three female governors. In the House of Representatives a big jump after the 1992 elections (the year of the woman) from 28 to 47 female representatives was followed by three elections that raised the number of women in the house by only nine to 56; women are still less than 13 percent of the 106^th Congress. In 1998 there were only three female congressional candidates for every seventeen male candidates.

As has been well documented, the deficit of female candidates in the U.S. owes much to the relatively small number of women who have, until recently, been willing or encouraged to run for office (Rule 1981, Darcy, Welch, and Clark 1994, Carroll 1994, Niven 1998). Women face three hurdles on their way to elected office: they have to want to run; parties have to select them as their candidates; and they have to win approval from the voters (Norris 1987).

This paper is about the second and third hurdles -- voter acceptance and the ways that party selectors view female candidates. We report the results of an experiment involving 820 randomly selected adults, and explore the conditions under which a female candidate are perceived more favorably than an otherwise identical male candidate. The experiment was funded by the Republican Network to Elect Women (RENEW) and conducted in late 1993 by Public Opinion Research, a professional survey organization based in Alexandria, Virginia. Though funded by a Republican support group, there is no partisan bias in the questions asked or the sample taken.

Our paper is distinctive in that, unlike virtually every other published experiment, this experiment provides information on a candidate’s party identification to test whether gender remains salient when the most prominent of political cues, party, is included in the description of a candidate. This is the only experiment of its kind using a large national randomized sample of adults, rather than college students. Furthermore, we test for a number of gender-based effects including affinity effects, wherein males prefer male candidates and females prefer female candidates. We proceed in three stages: we review 25 years of experiments contrasting male and female candidates; we test for party and gender affinity effects in the described experiment; and we discuss implications for increasing the presence of women in the U.S. Congress.

Survey research points to consistent differences in the public images of female and male politicians. Women are thought to be more emotional, more compassionate, better at dealing with family problems, weaker in times of crisis, weaker on crime, and so on (Hickman-Maslin Research 1987 cited in Burrell 1994 pp. 22-7, Simon & Danziger 1991, Cook & Wilcox 1995). These stereotypes are often perpetuated in the news media (Kahn 1996; Norris 1996), and there is some evidence that female legislators actually do a better job representing constituents on "social policy" issues (Dodson et al. 1995). To some extent, these perceptions reflect assumed trait differences between men and women in the population as a whole, and stereotypes are carried into the voting booth whether warranted or not (Dolan 1997, Koch 1999). All else equal, according to popular and scholarly surveys, women are at a disadvantage on several policy dimensions that matter greatly to voters, such as economics, national security, and leadership. However, "all else" is never really equal in comparisons among politicians. Every candidate, every district, and every electoral season is different in ways that are not easily captured through surveys. That is why our predecessors (and we) have opted for an experiment.

Experiments are powerful tools. The control they afford allows for precise estimates of how a variable affects outcomes in an experimental setting. Experiments designed to study the effects of candidate gender vary, but the general format is that a candidate is described -- or a speech is read -- and respondents are asked to rate the candidate on several issue dimensions and say whether or not the candidate would get their vote. What makes the process an experiment is that while everyone reads the same description or speech, one half of the sample, chosen at random, is informed the candidate is a woman while the other half is led to believe the candidate is a man. Since all other stimuli are exactly the same, and given that respondents are randomly placed into treatment and control groups, any statistically significant differences between the two sub-samples must be because candidate gender affects the way respondents assess the candidates.

There are a number of cognitive strategies that voters use to help them cope with the task of assessing political candidates (Popkin 1995). The experiments reviewed below are designed to provide information on the use of one of these heuristic tools, the schemas that are invoked by voters upon receiving a cue about a candidate’s gender or party identification. Schemas are road maps in our memory, based upon socialization and prior experiences. They help us process new information quickly by providing us with hypothesized connections between traits, behaviors, and beliefs (Fiske & Kinder 1981; Markus and Zajonc 1985; Schank & Abelson 1977). Those hypothesized connections may not be entirely correct for each case, but schemas provide a useful shorthand to help interpret actions and allow us to make judgments.

Schemas allow voters to go beyond the limited information they have to make broader assessments on a wide range of attributes and phenomenon. Previous research on schemas find that given a candidate’s party, voters are able to infer a candidate’s position on issues, the candidate’s placement on an ideological scale, and estimate their own willingness to support a candidate (Conover 1981; Conover and Feldman 1989; Feldman and Conover 1983; Rahn 1993). Rahn (1993) has run experiments showing that party schemas are so powerful that they can overwhelm issue information cues that are inconsistent with the party schema.

While the connection between party label and candidate position is a reasonably close inference jump, voters have been shown to make political judgments on weaker and less obviously relevant information. Rapoport, Metcalf, and Hartman (1989) found that while few respondents in their control group were willing to make inferences of policy positions for candidates when they had no information (15%), a substantially higher number (59%) inferred a candidate’s supposed policy positions based solely on personality traits. Our experiment tests to see if respondents make political inferences based on a candidate’s sex, even when party information is available, and how party schemas and gender schemas interact.

Since 1975, fourteen experiments aimed at understanding how gender affects the perceptions of political candidates have been reported in political science and social psychology journals. These experiments are summarized in a table in the Appendix. We cannot be certain that this is a complete census of all relevant articles, but we are confident that we have identified the bulk of articles.

The experiments can be sorted into two categories. The first set tests whether experimentally "changing" a candidate’s gender affects the votes for the candidate. Experiments in the second set of experiments focus on whether male or female candidates are thought to be more competent in dealing with specific policy areas. A sub-group of the second set also looks at the effect of traditionally "masculine" and "feminine" characteristics on estimates of competence (Rosenwasser et al. 1987, Rosenwasser & Seale 1988, Rosenwasser & Dean 1989).

For those studies assessing whether gender has a direct effect on the votes a candidate garners, the most common approach presents respondents with a hypothetical election and descriptions of two candidates. Respondents are then asked to mark a ballot (Adams 1975; Eckstrand & Eckert 1981; Garrett & Brooks 1987; Huddy & Terkildsen 1993a; Leeper 1991; Riggle et al 1997; Sapiro 1981-1982; Sigelman & Sigelman 1982; Spohn & Gillespie 1987). The first of these experiments, by Adams (1975) found that being female hurt candidates for the presidency in terms of votes, but helped city council candidates. The gender effects among U.S. senate and mayoral candidates were mixed. Gender stereotypes interacted with the level of the office sought. While Adams found differences in levels of support depending on the candidate gender, all other studies (8 of 9) fail to find any direct effects at the ballot box. Some studies found an affinity effect with female candidates doing slightly better with female voters and male candidates doing slightly better with male voters, but the effects tended to cancel each other out so that neither sex had an advantage in balloting.

A second approach, focusing on whether female or male candidates are thought to be better at handling issues in specific policy areas, has found much stronger gender effects. Typically, respondents are given a single candidate to evaluate on a questionnaire, based either on a short description (Huddy and Terkildsen 1993b) or a speech (Sapiro 1981-82, Leeper 1991). Again the candidate's gender is manipulated. Sapiro found that females are viewed as superior on education and health issues, while males are considered better on farm and military issues. Sapiro's research design has been modified and replicated several times. Subsequent studies show there are general categories of public policies (such as "nurturing" and "compassion" issues) in which female candidates are deemed better (Huddy & Terkildsen 1993b). "Education," "helping the poor," and "supporting the arts" are all issues that voters seem to believe female candidates are better at handling.

Concerning issues that male candidates are superior at handling, the experimental results are ambiguous. Neither Leeper (1991) nor Kahn (1992) report any issue areas in which male candidates have an assumed advantage in terms of competence. Sapiro (1981-1982) and Huddy and Terkildsen (1993b) find that being male provides a positive effect for presumed competence on military issues. The studies done by Rosenwasser and her colleagues (1987, 1988, 1989) find that male candidates are seen as stronger than female candidates on a "masculine task" scale that is dominated by military issues.

Several experiments have explored the effect of gender on how voters view a candidate’s character. Sapiro (1981-82), Leeper (1991), and Kahn (1992) all report that female candidates are seen as more honest or better able to maintain integrity in public office. Scholars have also tested whether traditionally male or female traits affect evaluations of candidates or legislative effectiveness (Rosenwasser & Dean 1989, Huddy & Terkildsen 1993a, 1993b). These experiments show that male traits are generally seen as more desirable and are associated with being a more effective legislator. What these studies fail to do, however, is prove that male candidates automatically have higher levels of these "male traits" -- such as being articulate. Regardless of gender, we expect most candidates to be articulate, so we are therefore reluctant to ascribe stereotypical male and female "traits" among the general public to the highly selective group of people who run for Congress.

We have two concerns about previous experiments. First, we raise a standard refrain, the "college sophomore problem." Thirteen of fourteen experiments we review use students as subjects, almost exclusively college undergraduates. The exception is Kahn's (1992) work on U.S. Senate races, and she uses adults from a single city, Ann Arbor, Michigan. As David Sears (1986) details, there are serious limitations to using college students in experiments. Sears (1986:522) notes late adolescents and young adults have less stable and more poorly crystallized social and political values than the public as a whole. Sears also notes their training and the format of psychological laboratory experiments lead to a very heavy emphasis on cognitive as opposed to affective bases of decision making. We know college students are, on average, more liberal on social issues, including the proper role for women in society, and come from higher socio-economic backgrounds. College students are also less likely to vote, their party identifications are more fluid, and they are less likely to have extensive experience with politicians, male or female. All these characteristics probably dampen the differential impact of a candidate's gender in college based laboratory experiments. Because university students are not typical of the national population on a host of factors, the external validity of studies using them is threatened.

Our second general concern is that several of the experiments have given respondents very little information with which to evaluate candidates, and this calls into question the generalizability of previous results. Most importantly, thirteen of the fourteen published experiments fail to include any information about a candidate's partisanship as a possible cue to voters, even though a candidate’s political party is the best predictor for what voters will do in the ballot booth (Wattenberg 1996, Miller and Shanks 1996). Rahn (1993:473) notes "the most powerful cue provided by the political environment is the candidate’s membership in a particular political party." For most voters, party identification amounts to a "standing decision" or a "brand loyalty." It is entirely possible that a voter’s party identification can overwhelm gender schemas. Unfortunately, because previous experiments have failed to build information on political party into their design, we do not know if this true.

The impoverishment of information in previous experiments has also led to contradictory results when scholars have tested for affinity effects (Ekstrand & Eckert 1981, Sapiro 1981-82, Sigelman & Sigelman 1982, Garrett & Brooks 1987, Spohn & Gillespie 1987, Riggle et al. 1997). A gender-based affinity effect is present when voters support candidates who are similar to themselves: men are more likely to vote for male candidates or women are more likely to vote for female candidates. One might expect to find an affinity effect in answers to statements such as "This candidate shares my concerns" or "This candidate cares about people like me." When those sentiments are turned into a vote for a specific candidate, however, male-male and female-female affinity effects can cancel each other out, leaving no real impact on who is elected. The most definitive non-experimental study of the affinity affect concludes plainly, "On average, women have been slightly more likely than men to vote for women candidates" (Seltzer et al 1997:102. See also Cook 1994; Cook & Wilcox 1995; Dolan 1998; Paolino 1995). Steckenrider and Thompson (1997), using a quasi-experimental design, also find a greater willingness for women to vote for women.

Five previous experiments explicitly tested for affinity effects. The two studies reporting pervasive affinity effects, however, provided respondents with very little substantive information to use in evaluating the candidates (Sigelman & Sigelman 1982, Garrett & Brooks 1987). Such "first impressions" based on gender likely have stereotypes doing most of the work. The three other studies provided respondents considerably more background on the candidates, and affinity effects disappeared under most model specifications (Ekstrand & Eckert 1981, Sapiro 1981-82, Riggle et al. 1997).

For many voters assessing candidates, the experimental descriptions are fairly realistic -- but for the absence of political party cues. Most voters are cognitive misers interested in the minimum amount of information needed to make a decision. This is especially true for evaluating elected offices farther down the ballot. Tests of information recall on candidates' names, positions on issues, and background for congressional candidates show that only a small portion of the electorate knows anything more than superficial information about candidates (Delli Carpini & Keeter 1996). Elections to the House of Representatives in particular have been identified as being low information elections (Abramowitz 1980; Hinckley 1981; McDermott 1997; Stokes and Miller 1966). It is precisely in this limited information environment that political party labels are most important, and on this the scholarly literature is clear: partisanship is the strongest predictor for how voters choose among congressional candidates.

Previous experiments have provided political scientists and social psychologists with several valuable insights. Voters view male and female candidates differently. There are also indications that male and female candidates are assessed differently depending on what issues are engaged in an election campaign. Experiments have come to these conclusions while giving respondents sketchy candidate descriptions. We do not take issue with this except to note that affinity effects and differential evaluations of expertise may fade as voters learn more about candidates. The relevance of these experiments are, however, strengthened by the fact that most studies of congressional elections show voters know relatively little about political candidates they vote for. They do, however, know the partisan affiliations of these candidates, and the failure of earlier experiments to include partisanship is a serious omission.

The experiment described below presents results from a national survey commissioned by the Republican Network to Elect Women (RENEW). A random sample of 820 adults from throughout the United States were polled December 6 - 8, 1993. The survey -- designed to conform to an experimental methodology -- was conducted by Public Opinion Research, one of the country's leading political polling firms. Each respondent was read the same candidate description. Half of the respondents were told that the candidate was a male, the other half were told that the candidate was a female. Each respondent heard the following:

I am going to read you a brief description of a potential candidate for Congress in this area. After I read this, I will ask you to evaluate [him/her].

The candidate is a Republican [man/woman] who has never run for office before, but has been active in the community. [She/He] is a businessperson who is running because [he/she] says that Congress "just doesn't get it" and wants to bring a common sense business approach to government. [His/Her] first priority is to work to reduce government spending and waste.

After the candidate was described, respondents were asked to evaluate the likelihood they would support the candidate. Respondents also evaluated the degree to which the (male or female) candidate possessed different traits (such as leadership qualities). Answers to those questions serve as dependent variables in the data analysis described below. Regarding this specific experimental design, the inclusion of "business skill" as one of the candidate’s strengths eliminates an expected male advantage. If financial acumen is considered a latent "male trait," then explicitly describing a woman as possessing this expertise would cause some of the assumed advantages for men to disappear. Unfortunately, because the experimental design only tested voter reactions to female and male Republicans it is impossible to discuss how gender might affect evaluations of Democratic candidates.

What do we expect? Our first concern is whether a candidate's gender affects the likelihood voters will support or oppose the candidate. We have argued that gender-based affinity effects and issue differences probably diminish in information-rich elections.
We expect that once a party label is added to a candidate that the candidate's gender will not affect how our respondents evaluate the candidate. Our presumption is that party is a much more powerful cue when evaluating candidates.

A second set of expectations involves perceived candidate characteristics. Social psychologists have shown that there are consistent gender stereotypes subscribed to by both men and women (Williams & Best 1990). Women are generally seen as more nurturing, more supportive, less confrontational, and in some cases more honest. Men, on the other hand, are thought to be more decisive, stronger, and better able to deal with crises.

An important question is whether these traits that are generally attributed to men and women are also attributed to political candidates of each gender. Some of these traits have been tested and found in previous experiments (Sapiro 1981-82, Rosenwasser & Dean 1989, Leeper 1991, Kahn 1992, Huddy & Terkildsen 1993a). While the survey we are exploring does not ask about the exact factors described in the gender stereotype research, it does explore a number of closely related issues. Based on previous research, we would assume that the more masculine factors -- such as being well qualified and a strong leader -- would favor the male candidate, while being able to "share my concerns" (which suggests an ability to empathize) and being "trustworthy" would favor the female candidate.

Our first analyses consider the effects of gender affinity and partisan affiliation. If there is a gender affinity effect, males should be more likely to vote for male candidates, and females should be more likely to vote for female candidates. The data indicate a slight, though not statistically significant (p=.25), affinity effect among male voters (Table not shown). 41.9% of male respondents who were told the candidate was male said they were very likely to vote for him, while 35.7% of males who were told the candidate was female were very likely to vote for her. The results show no such effect for female voters: 32.4% of female respondents would be very likely to vote for the male candidate, and 31.2% were very likely to vote for the female candidate.

Regarding partisan effects, Table 1 shows responses to the question asking if respondents were likely to support this candidate, broken down by party identification. Not surprisingly, party identification has a dramatic effect. Among Republican identifiers, 47.1 percent said they were very likely to support the candidate, while only 9.2 percent saw it as not very likely or not likely at all. On the other hand, only 19.6 percent of the Democrats said they were very likely to support the candidate while 36.1 percent saw it as not very likely or not likely at all. Independents held opinions between the two parties.

Source: Public Opinion Research, for RENEW, December 1993, N=820. Sample size in the table is 770 because "don’t know" and "refused to answer" responses are dropped.

While we found strong evidence of a party affiliation effect and little evidence of a direct affinity effect, we are especially interested in how a respondent’s party identification and a candidate’s gender interact. For example, do Democrats view female Republican candidates more favorably than otherwise identical male Republican candidates? In the first four rows of Table 2, we report the difference, based on gender, in the percentage of respondents who state the candidate is "very well" or "somewhat well" described by the indicated characteristics (as opposed to "not very well" or "not very well at all"). A positive percentage difference indicates the female candidate is advantaged over an otherwise identical male candidate.

* P < .10, ** P < .05, and *** P < .01 on a two-tailed test

Among all respondents, the voters consider the female Republican more trustworthy and more likely to share their concerns. The fifth row of Table 2 shows that on the tally that counts the most in democracies – voting – respondents prove more likely to vote for the female candidate. The percentage differences on these three questions are statistically significant, but what is especially noteworthy is how these differences break down across party groups. For example, among Democratic voters, the advantage for the female candidate on "trustworthiness" tops nineteen percent. Her advantage among both Democratic and Independent voters on empathy ("shares my concerns") hovers near ten percent. These two qualities -- identified in the social psychology literature as more typical of females overall -- have a direct impact at the ballot box (Kinder 1986). The female candidate's advantage in terms of likely support is ten percent among Independent and Democratic voters. In competitive elections, holding everything else constant, a ten-point swing from Independents and Democrats is something Republican party leaders should savor.

The impact of the candidate's gender on perceived leadership abilities differs markedly among the three partisan groups. Among Republicans, the female candidate is thought to be a weaker leader, while Independents and Democrats hold the reverse view. Why? We can only speculate. Perhaps the various party groups have different definitions of what it means to be a strong leader. Or perhaps Independents and Democrats assume that any woman slugging it out in the Republican party must have moxie. Whatever the reason for differences in the ways candidates are viewed, in terms of empathy, trust, and leadership, a connection to partisanship is unmistakable.

Tables 1 and 2 are relatively easy to interpret, but they may mask effects that are related to how candidate and respondent characteristics interact. We turn now to an ordered logit estimation to test the impact of candidate gender, gender affinity and party-id on voting and evaluations of House candidates. We use ordered logit because the dependent variables (candidate attributes) have four possible responses, "not very well at all", "not very well", "somewhat well" and "very well". We have run the models below testing for interactions between age, race, income and candidate gender and we have found these factors have no impact on the overall willingness to support female candidates. Accordingly, we model voter preferences as a function of gender (candidate and respondent), affinity, and party.

In Table 3, all the independent variables are either zeros or ones, making interpretations of the coefficients straightforward. The excluded category is Republicans, meaning that the coefficient estimate for the candidate's gender reflects how Republicans view the female relative to the male candidate.

The multivariate analysis gives greater clarity to issues raised above. First, we find no gender affinity effects. The interaction term measuring an affinity effect approaches statistical significance in none of the five models. To confirm these results, we tested for affinity effects on multiple subsets of the population, including working women, black women, wealthy and poor women, and so on. We also looked for affinity effects among male respondents. We found none. The experimental evidence purporting to show a gender affinity effect has always been problematic, and it has disappeared in previous experiments that provided relatively rich cues about candidates. We suspect that – in most real-world elections – voters do not flutter like moths to gender-matching candidates. A candidate’s partisanship is a better guiding light.

Not surprisingly, in Table 3 we see that Democrats and Independents are consistently less likely to think that the Republican candidate (regardless of the candidate’s gender) "shares my concerns," "is qualified," "can be trusted," or "is a strong leader." Republican candidates begin with a big disadvantage, from the perspective of Democrats and Independents.

Most strikingly, however, the Republican female candidate is consistently advantaged by her gender, again from the perspective of Democrats and Independents. This is measured by the two party and gender interaction terms. Here we see that Democrats report that the female Republican candidate is better qualified, more trustworthy, and a stronger leader than an otherwise identical male Republican candidate. Most importantly – from the perspective of the Republican leadership – Democrats and Independents are more likely to vote for the female Republican.

We asked earlier could gender ever overwhelm partisan predispositions? In terms of how Democrats view the qualifications of female Republicans, apparently so. As the democratic partisan/candidate gender interaction shows, Democrats view the female Republican as more qualified to hold office than the male Republican. The estimated effect (a coefficient estimate of +0.70) erases a Democrat’s partisan bias against Republican candidates in general (-0.70). On whether a candidate can be trusted, the partisan impact is again nearly erased (+1.10 to -1.16).

What do respondents mean when they say that a candidate is "qualified" for public office? We have a sense from the political science literature for features associated with successful candidates. Several of the characteristics we have already discussed -- such as trust, empathy, and leadership skills -- figure prominently in evaluations of which candidates voters think are "qualified" (Canon 1990, Fenno 1978). These qualities would seem to favor women, though another oft-cited dimension -- electoral experience -- hurts female candidates, because so few women have held office in the past (Seltzer et al 1997, Krasno & Green 1988).

Interestingly, the multivariate analysis fails to show a statistically significant advantage for women on one character trait (empathy) that is stereotypically thought to be to a female candidate's advantage. We do find, however, that both Democrats and Independents see female Republicans as stronger leaders than male Republicans. This sets up an interesting strategic dilemma for Republican women. In their primaries, leadership issues are unlikely to strengthen their candidacy, while leadership may play well as an issue in a general election.

That Democrats are more likely to trust a Republican female than a Republican male raises a minor puzzle. Trust in government and politicians has been steadily declining in the United States and most Western democracies (Nye, Zelikow & King 1997). It is possible that the female Republican candidate, as has been suggested of female candidates in general, are less likely to be seen as professional politicians. Our data support this assertion. In response to a question as to how well the term "typical politician" described this candidate, Independents and Democrats saw a clear distinction with the male candidate being seen as better described by this term. A candidate who conjures the image of the outsider untainted by the sordid goings-on in the halls of power is more convincing when the candidate looks different from those in the halls of power. Women candidates have that advantage, and it works to allay some of the initial skepticism Democrats and Independents have when they hear the candidate is a Republican. Kahn (1992) found a similar effect.

While trustworthiness, qualifications, and leadership skills are important, votes are paramount. Regardless of a candidate's gender, partisan predispositions are barriers that may block crossover votes. The "bottom line" in Table 3, however, confirms findings reported in Table 2. Female Republican candidates are more likely than male Republican candidates to get votes from Democrats and Independents. This positive gender effect for Republican women is only true for Democrats and Independents, however. As the negative coefficient on candidate gender indicates, Republicans are less likely to support a female candidate relative to an otherwise identical male candidate.

Implications

One implication from our results is of interest mainly to experimental social scientists: party identification provides a powerful cue in evaluating candidates. Partisan identification should be included in future experiments that consider how voters evaluate candidates, even if the primary interest is in questions of gender.

There is, however, a broader finding here, one that has been anecdotally told in the popular press (Abramson & Seib 1996). The Republican party, if it is interested in increasing its votes and seats, should run more female candidates. This is especially true in those districts where crossover votes from Independents and Democrats are necessary for Republican victories. Within the American system, however, decisions on who is to be the party’s candidates in congressional elections are not made by a rational vote-maximizing unitary actor; rather they are made in the messy and chaotic world of party primaries.

For the Republican party to be able to take advantage of a woman’s perceived strengths as a candidate, she must first win her party primary. The support that prospective female Republicans get from Democrats and Independents will matter little if they run into trouble in the primaries. Table 4 explores this concern by showing the percent of respondents "very likely" to vote for the male and female candidate across the standard seven point party identification scale. The results are unmistakable. While strong Democrats are much more supportive of a female Republican than a male Republican, strong Republicans are much less supportive of the female candidate. The Republican female candidate is at her biggest disadvantage among respondents who consider themselves "strong Republicans;" there is a 15 percent gap among these respondents. This is especially problematic since "strong" partisans are far more likely than others to contribute time and money to a campaign, and they are much more likely to vote in primaries. Given that strong party identifiers are the most active slice of voters in primaries, our experimental data suggest that female Republicans will have a more difficult time getting nominated.

Why should staunch Republicans be more skeptical of a female candidate? McDermott (1997) suggests that gender provides a low cost social information cue as to the ideological bent of a candidate. She provides some evidence that voters perceive female candidates as more liberal. In addition, she finds that Democratic women did better with liberals and worse with conservatives than Democratic males in House elections, although she finds no effects for Republican women.

The results in our experiment are consistent with McDermott’s analysis based on the National Election Studies. Our respondents were asked how well the term "conservative" described the candidate being evaluated. The results are instructive. For Democrats and Independents there was very little difference in evaluations. 32.3% of those evaluating the male candidate said the label "conservative" described the candidate "very well". Among those who evaluated the female candidate, 34.1% reported conservative described the candidate "very well", a trivial difference. For Republicans, however, while 42.4% of those who evaluated the female candidate said the label described her "very well", the equivalent response for the male candidate was more than 15 percent higher at 57.7%. In addition, for Republicans the evaluation of how "conservative" the candidate is directly affected the vote. 71 percent of those who said the conservative label described the candidate "very well" said they were "very likely" to vote for the candidate, while only 27 percent of those who did not believe the conservative label fit "very well" were "very likely" to vote for the candidate. Apparently, being a woman sends a cue to Republican voters, but not to Democratic or Independent voters, that the candidate is less conservative than the men she may have to compete against in a party primary.

To the extent that conservatives are active in primaries, the impression that female Republicans are more liberal than otherwise identical male candidates works against women trying to win votes in the Republican primaries. These difficulties with primary voters come in addition to difficulties female candidates are likely to have with party gatekeepers. In surveying local party chairs of both parties, Niven (1998: 57) found that "potential women candidates are subject to bias in recruitment that hinders the cause of electing more women to state legislatures and Congress." He presents evidence that indicate party chairs prefer male to female candidates.

An obvious question when we see these experimental results is whether there is any corroborating evidence in real-world primaries. For a first look at this question we turn to the results from House of Representative primary elections in the 1990s. As female incumbents are disproportionately Democratic and overwhelmingly returned (50 of 50 female House incumbents who ran for re-election in 1998 won) including incumbent races would bias the results in the Democrat’s direction. Therefore Table 6 looks only at open seat districts, where there was no incumbent from either party. Table 6 shows the number of districts where a woman ran and won the major party nomination for the two major parties. What Table 6 shows is there are far more women running in Democratic open seat primaries than in Republican open seat primaries. Furthermore Democratic voters have shown a higher propensity to choose a female candidate. In 53.2% of Democratic primaries in open seat districts, where a woman has run, the party’s voters have selected a female candidate. While for the Republicans, women have won in 38.3% of the districts where there was a woman running. The almost 15 point difference in win rates is statistically significant (p<.04, one-tailed test). In a word, the empirical data support the experimental data; female candidates are less likely to win the primary if they are running in the Republican primary than if they are running in the Democratic primary.

There is no such thing as a non-partisan election for the U.S. Congress. Candidates are not simply males or females, blacks or whites, country-folk or city-folk. They are partisan creatures, born of party primaries, vying for jobs in an intensely partisan institution. Unlike previous experiments, we have tried to take account of the central role party plays in elections. Our results find that even when the information in the environment is enriched by including party labels, gender effects still appear.

As such our results are consistent with the results of a quasi-experiment run by Thompson and Steckenrider (1997). In their study they ran hypothetical elections, collected from a mail survey, where the amount of information about candidates varied from one election to the next. They included (or consciously withheld) information on gender, candidate’s party, and position on one issue (abortion). They found that party and issue cues were overwhelmingly used as the most relevant cue to voting. There were a subsection of voters, however, who did vote for women candidates based on gender and contrary to their party affiliation. The number of respondents voting this way was not large, but in a tight congressional race the ability to attract independent and cross-over votes is crucial.

While the assertion of female advantage in congressional races is relatively new among political scientists and social psychologists, it infected professional political consultants a decade earlier and is now conventional wisdom on the campaign trail. "Voters trust women more than men, believe that women listen better than men, and look to women for new ideas," argues Cathy Allen, a Seattle political strategist (Gugliotta 1996, C1).

The experimental results reported here are supportive of these suggestions and as such they are part of a growing experimental literature suggesting that female candidates have some distinct advantages in congressional campaigns. Our study shows, however, that those advantages are mediated by partisanship. For Republicans evaluating a female Republican, gender appears to send a signal that the candidate is more liberal than a comparable Republican male. This may lead to Republican women having a harder time winning the party’s nomination. For Independent and Democrats, on the other hand, being female does not send a strong ideological signal; rather, it provides signals on a different set of characteristics. Specifically, Democratic and Independent voters are predisposed to view female Republicans as more likely to share their concerns, more trustworthy, and better leaders. Entire political campaigns can be built around such positive traits.

In 1986, 52 percent of the women in Congress were Republicans. Fourteen years later, only 31 percent of the women in Congress are Republicans. While there has been a significant increase in the number of women running and winning under the Democratic label, the increase among Republican women has been much more modest. The reason does not appear to lie with the voting public as a whole. Our results show that Republican women have some distinct advantages over Republican men as general election candidates. Republican women, however, appear to have significant problems within their own party and especially with the most activist elements in their party. If the Republican party were more willing to turn to women as candidates we would see an increase in women’s representation and possibly also an increase in Republican victories. Representative Nancy Pelosi (D-Ca), citing women candidates’ ability to "identify" with voter concerns, said, "in marginal districts, women candidates have an advantage" (Cohen 1998). Most of the turnover in House elections occurs in these marginal districts, so the potential for change can be exploited by turning to female candidates.

There is clearly room for more experiments of the type reported here. An obvious follow up with a national sample is to see if similar effects appear for Democratic women. It is plausible, for example, that being a woman is especially helpful for Republican women, because it helps soften the often tough and uncaring image associated with some Republican policies. For Democratic women the gains may be much smaller. Regardless, we encourage the use of experiments, imbedded within surveys. There are many more questions of candidate evaluation that can be answered with a representative sample through the use of experimental methodology.

Abramson, Jill & Gerald F. Seib. 1996. "GOP Faces Dissent of Women Who are Moderate, Affluent." The Wall Street Journal. July 17, 1996. A1.

Adams, William C. 1975. "Candidate Characteristics, Office of Election, and Voter Responses." Experimental Study of Politics. 4:79-91.

Bernstein, Robert A, 1997. "Might Women Now have the Edge? Open-Seat House Primaries" Women and Politics 17(2): 1-26.

Burrell, Barbara C. 1994. A Woman’s Place is in the House: Campaigning for Congress in the Feminist Era. Ann Arbor: University of Michigan Press.

Canon, David T. 1990. Actors, Athletes and Astronauts: Political Amateurs in the United States Congress. Chicago: University of Chicago Press.

Carroll, Susan J. 1994. Women as Candidates in American Politics. 2nd Edition. Bloomington, Indiana: Indiana University Press.

                    Niven, David. 1998. "Party Elites and Women Candidates: The Shape of Bias." Women & Politics                          19(2): 57-80.